反復的重複部推定を伴うスキャンマッチングにおける位置合わせ時の停留状態解析及びその回復

摘要

Iterative Overlapped Part Estimation is a method for generating highly accurate global maps for efficient wide-area exploration by an autonomous planetary rover in the unknown natural field. The basic concept of the method consists of two parts. First, estimating the overlapping parts in the terrain data and eliminating the un-overlapped part to increase the expectation of the overlapped part rate, then maximizing the accuracy of alignment vector via alignment technique. Secondly, iterating these processes sequentially until converge to an optimal value. There is an issue that alignment result cannot converge to an optimal value and remain in another value. We call this situation Stationary State. In this paper we analysis and propose a method to recover Stationary State in Alignment by Iterative Overlapped Part Estimation.%未知な環境下で活動するローバは周囲の環境情報を逐次計測しながら探査活動を行う.計測された地形データを繋ぎ合わせ，広域地図を生成する事が探査計画の効率化，科学的地形理解など様々な面で価値がある.ここで我々は移動ロボットの地図構築のための手法「反復重複部推定」を提案してきた.本論文では提案手法における位置合わせが局所最適値に陥る停留状態について解析及びその回復を議論する.